# Cloudflare Workers AI [[Cloudflare]] Workers AI is a serverless inference platform that runs open-source models (Llama, Mistral, Whisper, Stable Diffusion, BGE embeddings, and dozens more) on Cloudflare's GPU fleet, accessible from any [[Cloudflare Workers]] script via a binding or from outside via REST. For years the product was purely operational: take well-known open models, host them on GPUs at the edge, expose them under a uniform API, charge per token or per request. No quotas to request, no GPU provisioning, no batch endpoints to wrangle. That's no longer the whole story. On October 1, 2026, the Workers AI team shipped its first models trained in-house: [[Cloudflare Clef]] (27B) and Clef-flash (9B), two decision models with open weights under Apache 2.0. Most of the catalog is still third-party open models, but Cloudflare is now a model lab too (a small one). ## Why It Matters For AI features in app code, Workers AI removes the "where do I run the model" question. Embedding generation, summarization, transcription, image generation — all become single function calls from a Worker. Paired with [[Cloudflare Vectorize]] it forms a complete RAG stack with three bindings: AI, Vectorize, and your data store. ## Model Catalog (representative, October 2026) - **LLMs**: [[Kimi K2.6]] and Kimi K2.7 Code, [[GLM-5.2]] and GLM-5.3, DeepSeek V4 Flash and Pro, [[Gemma 4]] 26B A4B, Qwen 3.8 27B, gpt-oss 20B/120B, Nemotron 3 Super, Llama 3.3 70B and Llama 4 Scout - **Decision models**: [[Cloudflare Clef]] and Clef-flash, Cloudflare's own - **Embeddings and reranking**: BGE small/base/large/M3, EmbeddingGemma, Qwen3 embedding, BGE reranker - **Speech**: Whisper (transcription), Deepgram Nova 3 and Flux (speech-to-text), Deepgram Aura and MeloTTS (text-to-speech) - **Vision**: Llama 3.2 Vision and Moondream 3.1 (image understanding), FLUX.1 and FLUX.2, Leonardo Phoenix, Stable Diffusion XL (generation) - **Translation**: M2M-100, IndicTrans2 - **Classification, summarization, code completion**: rotating catalog The catalog rotates for real. On May 30, 2026, Cloudflare retired a batch of older models (Llama 2, Llama 3 and 3.1 8B/70B, Mistral 7B, Gemma 7B, Gemma 3 12B and more), and Kimi K2.5 requests were silently aliased to the pricier Kimi K2.6. Pin model IDs, and watch the changelog. ## Pricing Shape - Free tier: 10K neurons/day (sufficient for hobby projects) - Paid: $0.011 per 1K neurons (a "neuron" abstracts over token/image/second costs) - LoRA fine-tuning supported on a subset of models - Since July 28, 2026, the heavy models (Kimi K2.6/K2.7 Code, GLM-5.x, DeepSeek V4) need the Workers Paid plan or prepaid AI Gateway credits; Free plan calls get a 403 - Since August 7, 2026, Workers AI and [[Cloudflare AI Gateway]] share one entry point: `env.AI.run()` and the `/ai/` REST endpoints also reach third-party models, and AI Gateway credits can pay for Workers AI inference (unified billing) ## Common Use Cases - **In-Worker AI features** — summarize, classify, embed without leaving the request path - **RAG pipelines** with [[Cloudflare Vectorize]] - **Voice-to-text** workflows via Whisper - **Image generation** for thumbnails, avatars, content - **Cheap inference layer** in front of more expensive frontier models ## References - Workers AI home: https://ai.cloudflare.com/ - Model catalog: https://developers.cloudflare.com/workers-ai/models/ - Pricing: https://developers.cloudflare.com/workers-ai/platform/pricing/ - Changelog, Clef launch (2026-10-01): https://developers.cloudflare.com/changelog/post/2026-10-01-clef-workers-ai/ - Changelog, model deprecations (2026-05-08): https://developers.cloudflare.com/changelog/post/2026-05-08-planned-model-deprecations/ - Changelog, select models require Workers Paid (2026-07-28): https://developers.cloudflare.com/changelog/post/2026-07-28-models-require-workers-paid/ - Changelog, Workers AI and AI Gateway unify model access and billing (2026-08-07): https://developers.cloudflare.com/changelog/post/2026-08-07-workers-ai-unified-billing/ ## Related - [[Cloudflare]] - [[Cloudflare Workers]] - [[Cloudflare Vectorize]] - [[Cloudflare AI Search]] - [[Cloudflare Agents SDK]] - [[Cloudflare Sandbox SDK]] - [[Cloudflare Clef]] - [[Cloudflare AI Gateway]] - [[Wrangler]]